Antibiotic Use and New-Onset Inflammatory Bowel Disease in Olmsted County, Minnesota: A Population-Based Case-Control Study

Abstract

Background and Aims

Several studies have suggested significant associations between environmental factors and the risk of developing inflammatory bowel disease [IBD]. However, data supporting the role of antibiotics are conflicting. The aim of this study was to evaluate the association between antibiotic use and new-onset IBD.

Methods

We conducted a population-based case-control study using the Rochester Epidemiology Project of Olmsted County, Minnesota. We identified 736 county residents diagnosed with IBD between 1980 and 2010 who were matched to 1472 controls, based on age, sex and date of IBD diagnosis. Data on antibiotic use between 3 months and 5 years before IBD diagnosis were collected. Logistic regression models were used to estimate associations between antibiotic use and IBD, and were expressed as adjusted odds ratio [AOR] with 95% confidence interval [CI].

Results

Antibiotic use occurred in 455 IBD cases [61.8%] and 495 controls [33.6%] [p < 0.001]. In multivariate analysis, there were statistically significant associations between antibiotic use and new-onset IBD [AOR, 2.93; 95% CI, 2.40–3.58], Crohn’s disease [CD] [AOR, 3.01; 2.27–4.00] and ulcerative colitis [UC] [AOR, 2.94; 95% CI, 2.23–3.88]. A cumulative duration of antibiotic use ≥ 30 days had the strongest AOR [6.01; 95% CI, 4.34–8.45]. AOR for those receiving antibiotics under the age of 18 years was 4.27 [95% CI, 2.39–7.91], 2.97 for age 18–60 years [2.36–3.75] and 2.72 for age > 60 years [1.60–4.67].

Conclusions

This population-based case-control study suggests a strong association between antibiotic use and the risk of both new-onset CD and new-onset UC. The risk was increased among all age-onset IBD.

1. Introduction

Inflammatory bowel diseases [IBD], consisting of Crohn’s disease [CD] and ulcerative colitis [UC], are chronic idiopathic inflammatory conditions. Despite advances in the management of IBD, the pathogenesis of IBD is incompletely understood. The key contributors to the pathogenesis of IBD are thought to be genetics, host immune response and environmental factors.¹ The low concordance rate of IBD [approximately 35–50% for CD and 15% for UC] in identical twins suggests that genetic factors do not completely explain the disease aetiology.^2–4 Likewise, the increasing incidence of IBD worldwide including North America, Europe, Australia and Asia over the past 50 years suggests the important role of host immune response and environmental factors.^5–7

Emerging evidence suggests that disturbances in the gut microbiota and their interaction with the host immune system in a genetically susceptible individual may be the fundamental development of IBD.^1,8 Changes in microbial diversity and loss of protective microbial species have been demonstrated in patients with IBD.¹ In particular, some dietary changes and antibiotic exposure have been shown to alter gut microbiota.⁹ Several studies have suggested significant associations between environmental factors, including cigarette smoking^10,11and westernized diet,^12,13 and the risk of developing IBD. However, data for antibiotics are conflicting.^14–20 A previous meta-analysis has shown a modest association between antibiotic use and IBD development (pooled odds ratio [OR], 1.57; 95% confidence interval [CI], 1.27–1.94).¹⁵ Antibiotic use increased the odds of CD development [OR, 1.74; 95% CI, 1.35–2.23] but not UC [OR, 1.08; 95% CI, 0.91–1.27]. The authors noted a great deal of heterogeneity across the studies. For example, a study from the UK using the General Practice Research Database²⁰ and a study from Canada using the University of Manitoba IBD Database¹⁶ relied upon diagnostic and prescription codes to identify cases and exposures. As a result, misclassification bias may have occurred. Furthermore, some important predisposing factors to development of IBD, including cigarette smoking status and family history of IBD, could not be identified by administrative data. On the other hand, previous studies using a self-administered questionnaire^14,17–19 to assess antibiotic exposure were limited due to recall bias. Moreover, using prevalent patients may overrepresent those who regularly follow up, and may not include patients who migrated, lived in rural regions or those with indolent disease who were followed by their primary physician.¹⁹

Hence, the implications of antibiotic use as a predisposing factor in IBD aetiology remain unresolved, hindering our understanding of the pathogenesis of IBD. Therefore, to address this gap in knowledge, the present study sought to evaluate the association between the antibiotic use and new-onset IBD [both CD and UC] in a population-based cohort from Olmsted County, Minnesota.

2. Material and Methods

2.1. Rochester Epidemiology Project

The resources of the Rochester Epidemiology Project were used to identify permanent residents of Olmsted County, Minnesota, who were diagnosed with IBD and non-IBD between 1980 and 2010. The Rochester Epidemiology Project is a unique medical records linkage system developed in the 1950s and supported by the National Institutes of Health.²¹ It exploits the fact that virtually all the health care for residents of Olmsted County is provided by two organizations: the Mayo Medical Center, consisting of the Mayo Clinic and its two affiliated hospitals [Rochester Methodist and St. Mary’s], and Olmsted Medical Center, consisting of a smaller multispecialty clinic and its affiliated hospital [Olmsted Community Hospital]. In any 4-year period, over 95% of county residents are examined at either one of the two healthcare systems.²² Diagnoses generated from all outpatients visits, emergency room visits, hospitalizations, nursing home visits, laboratory, surgical procedures, medical treatment, autopsy examination and death certification are contained in a single comprehensive medical records linkage system. Thus, this population-based data resource ensures virtually complete case ascertainment and follow-up in a well-defined geographical region.²³

2.2. Study population

This population-based case-control study comprised 736 Olmsted County residents first diagnosed with IBD [339 CD and 397 UC] between 1980 and 2010 according to well-defined criteria as previously described.^5,24–26 The index date was based on the first IBD diagnosis. For each case with IBD diagnosis, two controls were randomly selected from Olmsted County residents without an IBD diagnosis after matching for age, sex and index date of IBD diagnosis, which assigned the index date corresponding to the index date of IBD diagnosis. To determine whether IBD developed or not, individuals in the control cohort were followed through their medical history from the index date until death, emigration or the end of the study [June 30, 2016], whichever came first. We investigated the impact of age on the association between antibiotic use and IBD development. Age groups at index date were stratified as <18 years, 18–60 years and > 60 years.

2.3. Antibiotic use

Data on the use of antibiotics with regard to class of antibiotic, indication, duration and number of antibiotic courses were collected for 5 years before the index date. Because antibiotics are commonly prescribed for symptoms of IBD prior to diagnosis, antibiotic use within 3 months prior to the index date was excluded. Classes of antibiotic were categorized as follows: penicillins, macrolides, cephalosporins, sulfonamides, quinolones, tetracyclines, metronidazole and others. Indications for antibiotic use were classified as: [i] ear, nose, throat and respiratory tract infection; [ii] gastrointestinal tract infection; [iii] genitourinary tract infection; [iv] skin, soft tissue and bone infection; or [v] other organ or systemic infection. The number of antibiotic courses was categorized as none, 1–3, 4–6, and ≥ 7. The start and stop dates of antibiotic use were recorded. The cumulative duration of antibiotic use for each individual was categorized as none, 1–14 days, 15–29 days and ≥ 30 days.

2.4. Statistical analysis

Assuming the prevalence of antibiotic use in the control group ranged from 25 to 50%, a sample size of at least 330 cases and 660 controls in each CD and UC cohort was required to detect an OR of at least 2 with a power of 80% at a two-sided significance level of 0.05. Descriptive statistical analysis was used to determine baseline characteristics of IBD cases and controls. To compare categorical variables, a chi-square test or Fisher exact test was used. To compare continuous variables, Student’s t-test or Mann–Whitney U-test was used. Antibiotic use and potential confounding factors (i.e. first degree family history of IBD, current cigarette smoking, former cigarette smoking, obesity [body mass index ≥ 30 kg/m²], daily aspirin use and statins use) were tested by univariate analysis. Daily aspirin use was defined as taking aspirin at least once a day for at least 3 months before the index date. Statin use was defined as taking a statin [pravastatin, atorvastatin, simvastatin, rosuvastatin, fluvastatin, lovastatin] for at least 3 months before the index date. A multivariate analysis model was then analysed using unconditional logistic regression models to estimate the adjusted OR [AOR] and 95% CI. Age, sex and any variables with a p < 0.1 in the univariate analysis were included in the multivariate analysis. Separate analyses for IBD subtypes [CD and UC] and age groups at the index date were performed. To evaluate whether there was a dose-dependent relationship between antibiotic use and the risk of IBD development, the analysis was also stratified by the cumulative duration of antibiotic use and by the number of antibiotic courses. An alpha-level of 0.05 was considered as statistically significant. Statistical analyses were performed by using the JMP 10 statistical software package [SAS Institute Inc., Cary, NC, USA].

3. Results

A total of 736 incident IBD cases and 1472 controls were included in this study. This sample size was sufficient to achieve the a priori statistical power assumptions for the study. Of the IBD cases, 339 had CD [46%] and 397 had UC [54%]. IBD cases had data available with a median duration of 5 years before the index date [range, 0.5–5 years] and 14.8 years after the index date [range, 0.1–36.2 years]. Controls had data available with a median duration of 5 years before the index date [range, 0.4–5 years] and 10 years after the index date [range, 0.1–36.5 years]. Table 1 shows the demographic characteristics of the IBD cases and controls. By study design, there were no differences in the age and sex distribution between IBD cases and controls. In both IBD cases and controls, the median age at the index date was 34 years [range, 1.2–93 years], and 55% were male.

Table 1.

Demographic characteristics of IBD cases and controls at the index date from 1980 to 2010

Characteristic	IBD [n = 736]	Controls [n = 1472]	p-value
Median age at index date, years [range]	34 [1.2–93]	34 [1.2–93]	1.00
< 18 years, n [%]	87 [11.8%]	174 [11.8%]
18–60 years, n [%]	559 [76.0%]	1118 [76.0%]
> 60 years, n [%]	90 [12.2%]	180 [12.2%]
Male, n [%]	405 [55%]	810 [55%]	1.00
Race, n [%]
White	665 [90%]	1074 [73%]	<0.001
African American	12 [1.7%]	81 [5.5%]
Asian	10 [1.4%]	52 [3.5%]
Other	6 [0.8%]	31 [2%]
Unknown	43 [6.1%]	234 [16%]
Ethnicity, n [%]
Hispanic	6 [1%]	45 [3%]	<0.001
Non-Hispanic	626 [85%]	996 [68%]
Unknown	104 [14%]	431 [29%]
Median duration before index date, years [range]	5 [0.5–5]	5 [0.4–5]	<0.001
Median duration of follow-up after index date, years [range]	14.8 [0.1–36.2]	10.0 [0.1–36.5]	<0.001
Obesity [BMI ≥ 30 kg/m2], n [%]	221/730 [30.0%]	371/1467 [25.2%]	0.08
1st degree family history of IBD, n [%]	42 [5.7%]	3 [0.2%]	<0.001
Smoking, n [%]
Current smoking	123 [16.7%]	229 [15.6%]	0.48
Former smoking	150 [20.4%]	152 [10.3%]	<0.001
Any antibiotic use, n [%]	455 [61.8%]	495 [33.6%]	<0.001
1–3 courses of antibiotics	357 [48.5%]	436 [29.6%]
4–6 courses of antibiotics	77 [10.5%]	53 [3.6%]
≥ 7 courses of antibiotics	21 [2.8%]	6 [0.4%]
Daily aspirin use, n [%]	33 [4.5%]	59 [4.0%]	0.60
Statin use, n [%]	20 [2.7%]	63 [4.3%]	0.08

IBD, inflammatory bowel disease; BMI, body mass index.

The use of antibiotics between 3 months and 5 years before the index date occurred in 455 IBD cases [61.8%] and 495 controls [33.6%] [p < 0.001]. Among the antibiotic users, IBD cases received a greater number of antibiotic courses [median, 2: range, 1–12] than controls [median, 1; range, 1–8] [p < 0.001]. There were no differences in indications for antibiotic use between IBD cases and controls with regard to: ear/nose/throat and respiratory tract infections [76% vs. 70%, p = 0.06]; skin/soft tissue and bone infections [25% vs. 24%, p = 0.29]; and genitourinary tract infections [23% vs. 23%, p = 0.96]. However, the indication of antibiotic use for gastrointestinal tract infections in IBD cases was higher than controls [6% vs. 3%, p = 0.01]. Regardless of antibiotic use, gastrointestinal tract infections occurred in 29 of 736 IBD cases [3.9%] and 14 of 1472 controls [1.0%] [p < 0.001]. Of these, Clostridium difficile infection occurred in two controls but none was found in IBD cases.

Table 2 shows the univariate and multivariate analysis for the association between various factors and new-onset IBD. In multivariate logistic regression, after adjusting for age, sex, smoking status, family history of IBD, obesity and statin use, there were statistically significant associations between any antibiotic use and the onset of IBD [AOR, 2.93; 95% CI, 2.40–3.58], CD [AOR, 3.01; 2.27–4.00] and UC [AOR, 2.94; 2.23–3.88]. These significant associations were observed across different antibiotic classes, including penicillins, macrolides, cephalosporins, sulfonamides, quinolones and tetracyclines, with AORs ranging from 2.01 to 2.53 for IBD, 2.21 to 2.81 for CD and 1.90 to 2.66 for UC [Tables 3 and 4]. Metronidazole was significantly associated with the risk of developing CD [AOR, 2.80; 95% CI, 1.14–7.11] but not with IBD overall [AOR, 1.84; 0.99–3.38] or UC [AOR, 1.38; 0.58–3.12]. Notably, three CD patients were given metronidazole for perianal infection prior to a definite CD diagnosis. After we excluded those cases, no statistically significant association between metronidazole and the risk of developing CD was observed [AOR, 1.94; 95% CI, 0.73–5.14].

Table 2.

Univariate and multivariate analysis of associations between various factors and new-onset IBD

Variable	Inflammatory bowel disease		Unadjusted			Adjusted†
	Cases [n = 736]	Controls [n = 1472]	OR	95% CI	p-value	OR	95% CI	p-value
Age [per 1 year]			1.00	1.00–1.00	1.00	1.00	0.99–1.00	0.16
Male [n]	405	810	1.00	0.84–1.19	1.00	1.24	1.01–1.51	0.04
Current smoking [n]	123	229	0.92	0.72–1.17	0.48	–	–	–
Former smoking [n]	150	152	2.22	1.74–2.84	<0.001	2.18	1.64–2.90	<0.001
1st degree family history of IBD [n]	42	3	29.6	9.15–95.9	<0.001	27.3	9.58–115.2	<0.001
Obesity [n]	221	371	1.20	0.98–1.47	0.08	1.14	0.92–1.42	0.22
Statin use [n]	20	63	0.62	0.37–1.04	0.08	0.46	0.25–0.80	0.01
Aspirin use [n]	33	59	1.12	0.73–1.74	0.60	–	–	–
Antibiotic use [n]	455	495	3.20	2.66–3.84	<0.001	2.93	2.40–3.58	<0.001

OR, odds ratio; CI, confidence interval.

†Age, sex, and any variables with a p < 0.1 in the univariate analysis were included in the multivariate analysis.

Table 3.

Unadjusted and adjusted odds ratio for IBD stratified by class of antibiotic

Antibiotic use	Inflammatory bowel disease		Unadjusted		Adjusted
	Cases [n = 736]	Controls [n = 1472]	OR	95% CI	OR	95% CI
Any antibiotic [n]	455	495	3.20	2.66–3.84	2.93	2.40–3.58
Penicillin [n]	246	228	2.74	2.22–3.37	2.53	2.02–3.18
Macrolide [n]	178	179	2.30	1.83–2.90	2.01	1.57–2.57
Cephalosporin [n]	107	94	2.49	1.86–3.34	2.29	1.68–3.12
Sulfonamide [n]	84	80	2.24	1.63–3.09	2.20	1.56–3.11
Quinolone [n]	75	73	2.17	1.56–3.04	2.10	1.47–3.01
Tetracycline [n]	55	46	2.50	1.67–3.74	2.10	1.37–3.22
Metronidazole [n]	22	26	1.71	0.96–3.04	1.84	0.99–3.38

IBD, inflammatory bowel disease; OR, odds ratio; CI, confidence interval.

Table 4.

Unadjusted and adjusted odds ratio for Crohn’s disease and ulcerative colitis stratified by class of antibiotic

Antibiotic use	Crohn’s disease		Unadjusted		Adjusted		Ulcerative colitis		Unadjusted		Adjusted
	Cases [n = 339]	Controls [n = 678]	OR	95% CI	OR	95% CI	Cases [n = 397]	Controls [n = 794]	OR	95% CI	OR	95% CI
Any antibiotic [n]	212	241	3.03	2.31–3.97	3.01	2.27–4.00	243	254	3.35	2.61–4.31	2.94	2.23–3.88
Penicillin [n]	119	117	2.59	1.92–3.50	2.66	1.94–3.65	127	111	2.89	2.16–3.87	2.42	1.76–3.33
Macrolide [n]	87	88	2.31	1.66–3.22	2.28	1.61–3.21	91	91	2.30	1.67–3.16	1.90	1.33–2.71
Cephalosporin [n]	51	53	2.09	1.39–3.14	2.21	1.44–3.37	56	41	3.02	1.98–4.60	2.66	1.71–4.16
Sulfonamide [n]	40	30	2.89	1.77–4.73	2.81	1.67–4.75	44	50	1.85	1.21–2.84	2.00	1.26–3.17
Quinolone [n]	35	32	2.32	1.41–3.83	2.41	1.41–4.11	40	41	2.06	1.31–3.24	1.94	1.18–3.16
Tetracycline [n]	24	20	2.51	1.36–4.61	2.34	1.24–4.43	31	26	2.50	1.46–4.28	2.10	1.18–3.76
Metronidazole [n]	12	9	2.73	1.14–6.54	2.80	1.14–7.11	10	17	1.18	0.54–2.60	1.38	0.58–3.12

OR, odds ratio; CI, confidence interval.

Subgroup analysis stratified by age group at IBD diagnosis showed that the association between any antibiotic use and the risk of IBD was statistically significant in all age groups. Those under 18 years receiving any antibiotic had an AOR for IBD of 4.27 [95% CI, 2.39–7.91], and the corresponding values at ages 18–60 years and > 60 years were 2.97 [2.36–3.75] and 2.72 [1.60–4.67], respectively.

A dose-dependent relationship between antibiotic use and the risk of IBD development was observed. Stratifying the analysis by the number of courses of antibiotic use, the AOR for one to three courses was 2.61 [95% CI, 2.12–3.23], 4.61 for for to six courses [95% CI, 3.12–6.84] and 10.34 for seven or more courses [95% CI, 4.19–29.19]. When the analysis was stratified by the cumulative duration of antibiotic use, there were significant associations with the risk of IBD. The AOR for 1–14 days was 2.27 [95% CI, 1.78–2.89] and 2.64 for 15–29 days [95% CI, 1.94–3.61], while using antibiotics ≥ 30 days had the strongest AOR of 6.01 [95% CI, 4.34–8.45]. We also found evidence of dose-dependent relationships between antibiotic use and the risks of developing CD or UC, as shown in Table 5.

Table 5.

Adjusted odds ratio for IBD, CD and UC stratified by course, duration and year before diagnosis of antibiotic use

Antibiotic use	IBD				CD				UC
	Cases [n = 736]	Controls [n = 1472]	Adjusted OR	95% CI	Cases [n = 339]	Controls [n = 678]	Adjusted OR	95% CI	Cases [n = 397]	Controls [n = 794]	Adjusted OR	95% CI
No use [n]	281	977	Ref		127	437	Ref		154	540	Ref
Any use [n]	455	495	2.93	2.40–3.58	212	241	3.01	2.27–4.00	243	254	2.94	2.23–3.88
Courses
1–3 courses [n]	357	436	2.61	2.12–3.23	160	206	2.65	1.97–3.58	197	230	2.65	1.99–3.53
4–6 courses [n]	77	53	4.61	3.12–6.84	38	31	4.37	2.58–7.45	39	22	5.23	2.9 2–9.56
≥ 7 courses [n]	21	6	10.34	4.19–29.19	14	4	11.89	4.00–43.6	7	2	10.88	2.49–75.17
Duration†
1–14 days [n]	208	293	2.27	1.78–2.89	91	136	2.17	1.53–3.07	117	157	2.30	1.66–3.18
15–29 days [n]	19	130	2.64	1.94–3.61	51	62	2.98	1.92–4.59	58	68	2.66	1.73–4.08
≥ 30 days [n]	138	72	6.03	4.34–8.45	70	43	5.89	3.80–9.23	68	29	7.07	4.31–11.83
Year before diagnosis
1 year [n]	187	174	3.60	2.75–4.72	91	92	3.32	2.30–4.81	96	82	3.65	2.50–5.35
2 years [n]	101	140	2.44	1.79–3.30	50	67	2.74	1.78–4.21	51	73	2.37	1.54–3.63
3 years [n]	71	79	2.62	1.79–3.81	33	34	2.98	1.73–5.14	38	45	2.75	1.65–4.57
4 years [n]	53	51	3.00	1.94–4.63	20	26	2.69	1.41–5.06	33	25	3.26	1.80–5.94
> 4 years [n]	43	51	2.60	1.63–4.11	18	22	2.93	1.48–5.72	25	29	2.42	1.31–4.43

IBD, inflammatory bowel disease; CD, Crohn’s disease; UC, ulcerative colitis; OR, odds ratio; CI, confidence interval; Ref, reference.

†Cumulative duration of antibiotic use 3 months to 5 years before diagnosis.

The significant association between antibiotic use and the risk of IBD remained regardless of year of antibiotic use before IBD diagnosis. For each individual year, the associations were statistically significant with AORs of 3.60 [95% CI, 2.75–4.72] at 1 year, 2.44 [1.79–3.30] at 2 years, 2.62 [1.79–3.81] at 3 years, 3.00 [1.94–4.63] at 4 years and 2.60 [1.63–4.11] at > 4 years before IBD diagnosis. These associations were observed in both CD and UC as well [Table 5].

4. Discussion

The results of this population-based case-control study demonstrate a strong association between antibiotic use and the risk of either new-onset CD or new-onset UC. In addition, significant dose-dependent relationships between antibiotic use and the development of CD and UC were observed. This relationship existed irrespective of age groups and presented across 5 years before IBD diagnosis. The majority of antibiotic classes were associated with an increased risk of developing CD and UC.

Our findings corroborate previous studies from the UK²⁰ and Canada.^16,27 Card et al. used the General Practice Research Database and illustrated the association between antibiotic use and the development of CD.²⁰ They showed that antibiotic use increased the odds of CD by a factor of over 30% [AOR, 1.32; 95% CI, 1.05–1.65]. Shaw et al. performed a case-control study using the University of Manitoba IBD Epidemiology Database.¹⁶ The authors found that the odds of IBD for those receiving antibiotic dispensation at 2–5 years before diagnosis was 1.29 [95% CI, 1.18–1.40] for CD and 1.26 [1.16–1.36] for UC. A dose-dependent relationship was shown. At 2 years before diagnosis, those receiving one or more and two or more antibiotic dispensations had a 1.27- and 1.48-fold increase in the odds of CD, respectively. The AOR of UC was 1.62 for three of more antibiotic dispensations. When stratifying by age groups, significant associations between antibiotic use within 2–5 years and adult-onset IBD were identified [AOR 1.32 for age 19–65 years and 1.25 for > 65 years].¹⁶ In a Manitoba nested case-control study, antibiotic use in the first year of life was associated with a 2.9-fold increased risk of new-onset IBD in childhood [age < 16 years].²⁷ Despite the use of large administrative databases, studies that rely on drug dispensation codes have limitations, in particular confounding factors. Important potential confounding factors such as family history of IBD and cigarette smoking status use were unavailable. Furthermore, the database could not verify that a dispensed antibiotic was actually consumed.

A key strength of this study is that all incident IBD cases and controls were derived from a well-defined geographical region and mostly received the same healthcare provision. More than 95% of Olmsted County residents receive medical care from either Mayo Clinic Medical Center or Olmsted Medical Center. Because of the medical record linkage system [Rochester Epidemiology Project], which provides access to all local medical records, we were able to review data from all available patients from 5 years before the index date until the end of the study period. We were able to verify the data regarding antibiotic consumption and accounted for other confounding factors. However, this study also had some limitations. First, because of its retrospective nature, any data not recorded in the medical records would have been missed. Secondly, socioeconomic factors are associated with hygiene status and healthcare utilization. The socioeconomic status of Olmsted County residents is slightly higher than that of the Upper Midwest population and the entire US population.²³ In 2000, 91% of Olmsted County residents were highschool graduates compared to 86% of the Upper Midwest population and 80% of the entire US population.²³ Median household income among Olmsted County residents [$51316] was higher than that of the Upper Midwest population [$43200] and of the entire US population [$41994].²³ The relatively high socioeconomic status in the region may limit the generalizability of these data to populations with different socioeconomic characteristics. Thirdly, Olmsted County had less ethnic diversity than the US population [90% vs. 75% white] between 1970 and 2000.²³ However, this gap has narrowed: 85.7% of county residents were whites in 2010 compared with 74.8% of the US population.²⁸ This change is similar to other Western countries. For example, according to the Office for National Statistics of the United Kingdom, the percentage of whites in England and Wales decreased from 94.1% in 1991 to 86% in 2011.²⁹ Although there were changes in the ethnic diversity of the Olmsted County population, the generalization of these study results to populations with more heterogeneous ethnicity may not be applicable.

It is unclear why antibiotic use was associated with new-onset IBD. Our findings may support a role of microbiome alterations in the pathogenesis of IBD. Previous studies in healthy humans have suggested that antibiotic exposure perturbs the gut microbiota composition,^30,31 and repeated antibiotic exposure results in persistent dysbiosis.^32,33 Many studies consistently reported changes in the total numbers and diversity of the gut microbiota in IBD patients.³⁴ For instance, increases in the numbers of adherent-invasive Escherichia coli, Enterobacteriaceae and Fusobacteriaceae, and decreases in the numbers of Bacteroides, Bifidobacteriaceae and Clostridia have been reported in IBD patients.¹ In addition to quantitative changes in the bacteria comprising the gut microbiota, antibiotics may alter the functional composition of the gut microbiota, as well.³⁵ Recently, a longitudinal paediatric IBD cohort study has shown an increased gut microbial dysbiosis index in new-onset IBD patients when compared to healthy controls.³⁶ In CD patients, the dysbiosis index at diagnosis was predictive of subsequent disease severity defined by the Pediatric CD Activity Index.³⁶

Although we found a strong association and dose–response relationship between antibiotic use and the development of IBD, a possible triggering role of antibiotics in the onset of IBD should be interpreted with caution. First, not all classes of antibiotics have the same local effect on the gut microbiome. Metronidazole, which directly targets specific gut bacteria, showed the strongest association with the approximately three-fold increased risk for developing IBD in the Manitoba study.¹⁶ In contrast to the Manitoba study, an increased risk for IBD in metronidazole was not observed in our study. The relatively small number of metronidazole users in our study may have contributed to the lack of statistical significance. For other classes of antibiotics, an approximately two-fold increased risk of IBD among each class of antibiotic [i.e. penicillins, cephalosporins, quinolones, sulfonamides, macrolides and tetracyclines.] was observed. Our results are line with previous studies. The Manitoba study showed significant associations with all types of antibiotics except clindamycin, with AORs of 1.12–2.86.¹⁶ In a large retrospective cohort study from the UK, tetracycline use for acne therapy was associated with IBD development (hazard ratio [HR], 1.39; 95% CI, 1.02–1.90].³⁷ Secondly, we examined antibiotic usage for 5 years prior to IBD diagnosis. We did not examine earlier periods or during the first year of life, which theoretically is the important period for development and adaptation of the gut microbiota.¹ Thirdly, antibiotic use may act as a surrogate marker for infection, leading to IBD development. A previous study found that the HR of developing IBD was 2.4 [95% CI, 1.7–3.3] in patients with acute gastroenteritis.³⁸ Hypothetically, either gastrointestinal tract infection or antibiotic use could be confounding factors. However, we found that only 3.9% of the IBD cases had a gastrointestinal tract infection. More than 70% of infections were ear, nose, throat and respiratory tract infections. These infection sites are concordant with a previous study of the association between antibiotic use in the first year of life and childhood-onset IBD.²⁷

In conclusion, antibiotic use was an independent risk factor for the development of both CD and UC in this population-based nested case-control study. The risk was increased among all age-onset of IBD. Further longitudinal prospective studies investigating the infection and antibiotic use in parallel with gut microbiome analysis in IBD are warranted to unravel the pathogenesis of IBD.

Funding

This work was supported in part by the Mayo Foundation for Medical Education & Research, and the Rochester Epidemiology Project [grant number R01 AG034676 from the National Institute on Aging of the National Institutes of Health]. The contents of the publication are solely the responsibility of the authors and do not necessarily represent the official view of the National Institutes of Health.

Conflict of Interest

None of the authors has any relevant conflicts of interests.

Author Contributions

SA: writing the proposal, submitting the proposal to Institutional Review Board, collecting the data, analysing the data, and drafting of the manuscript. WJT, LER and SVK: critical revision of the manuscript. EVL: conception and design of the study, critical revision of the manuscript. All authors provided final approval of the version to be submitted.